Article
Engineering, Electrical & Electronic
Ria Kanjilal, Muhammed F. Kucuk, Ismail Uysal
Summary: Human activity recognition (HAR) is an active area of sensory healthcare research that has the potential to improve the quality of life for patients and the public. This study investigates the impact of outlier users on activity recognition and proposes a novel approach called subtransfer learning to improve accuracy on challenging datasets with diverse users and sensor locations.
IEEE SENSORS JOURNAL
(2023)
Article
Computer Science, Information Systems
Tatsuhito Hasegawa, Kazuma Kondo
Summary: Sensor-based human activity recognition (HAR) is an important technology in IoT services. HAR using representation learning is widely used due to the difficulty of extracting meaningful features from raw sensor data. This study proposes an easy ensemble (EE) for HAR, which allows deep ensemble learning in a single model. Various techniques (input variationer, stepwise ensemble, and channel shuffle) for the EE are also introduced. Experiments on a benchmark data set demonstrate the effectiveness of EE and its techniques compared with conventional ensemble learning methods.
IEEE INTERNET OF THINGS JOURNAL
(2023)
Article
Computer Science, Artificial Intelligence
Wheidima Carneiro de Melo, Eric Granger, Abdenour Hadid
Summary: This article introduces a novel 3D CNN architecture (MSN) for effectively representing facial information related to depressive behaviors from videos. Experimental results show that the MSN architecture outperforms state-of-the-art methods in automatic depression recognition.
IEEE TRANSACTIONS ON AFFECTIVE COMPUTING
(2022)
Article
Computer Science, Artificial Intelligence
Chaolei Han, Lei Zhang, Yin Tang, Wenbo Huang, Fuhong Min, Jun He
Summary: This research focuses on enhancing the application of vanilla convolution in human activity recognition without adjusting the model architectures. Inspired by grouped convolution, a novel heterogeneous convolution is proposed to augment the receptive field of sensor signals for activity recognition. Experimental results demonstrate significant improvements in the performance of sensor-based activity recognition models.
EXPERT SYSTEMS WITH APPLICATIONS
(2022)
Article
Engineering, Electrical & Electronic
Ying Li, Junsheng Wu, Weigang Li, Aiqing Fang, Wei Dong
Summary: The sensor-based human activity recognition (SHAR) task aims to recognize signals collected by sensors in intelligent devices to assist people in their daily lives. Deep learning is being studied for combining with SHAR. To address the challenge of maintaining efficiency, an effective sensor signal representation method, called the temporal-spatial dynamic convolutional network, is presented. Extensive experiments demonstrate the superiority of this method over deep learning baselines and existing SHAR works on benchmark SHAR datasets.
IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT
(2023)
Article
Computer Science, Information Systems
Ling Pei, Songpengcheng Xia, Lei Chu, Fanyi Xiao, Qi Wu, Wenxian Yu, Robert Qiu
Summary: The article discusses using wearable inertial measurement units for human activity recognition with a focus on deep learning methods. It proposes building a large dataset and introducing a multi-domain deep learning framework to address technical issues. Experimental results demonstrate the efficiency and effectiveness of the proposed method.
IEEE INTERNET OF THINGS JOURNAL
(2021)
Article
Engineering, Electrical & Electronic
Sahil Waqar, Muhammad Muaaz, Matthias Patzold
Summary: Modern monostatic radar-based HAR systems perform well in detecting human activities towards or away from the radar, but fail to classify multidirectional activities. In this article, a distributed MIMO radar configuration is proposed to overcome this limitation by capturing and analyzing multidirectional human movements from multiple perspectives.
IEEE SENSORS JOURNAL
(2023)
Article
Computer Science, Artificial Intelligence
Chixuan Wei, Zhihai Wang, Jidong Yuan, Xiaokang Wang, Haiyang Liu, Qiyang Zhao
Summary: SemiHAR is a semisupervised human activity recognition method based on multitask learning, which addresses the challenges in activity recognition through generating 2D activity data and learning task relations.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2023)
Article
Chemistry, Analytical
Debarshi Bhattacharya, Deepak Sharma, Wonjoon Kim, Muhammad Fazal Ijaz, Pawan Kumar Singh
Summary: This study proposes an ensemble of deep learning-based classification models for predicting human activity, achieving high accuracy in biomedical measurements.
Article
Engineering, Electrical & Electronic
Hazar Zilelioglu, Ghazaleh Khodabandelou, Abdelghani Chibani, Yacine Amirat
Summary: This article proposes an alternative framework for semisupervised generative adversarial networks (GANs) using temporal convolutions for semisupervised action recognition in the HAR context. The framework addresses several problems related to conventional approaches, such as high dimensionality, scarcity of annotated data, scalability, and robustness. The effectiveness of the framework is evaluated on different datasets and shows high classification performance and generalization ability.
IEEE SENSORS JOURNAL
(2023)
Article
Engineering, Electrical & Electronic
Yang Wang, Hongji Xu, Yunxia Liu, Mengmeng Wang, Yuhao Wang, Yang Yang, Shuang Zhou, Jiaqi Zeng, Jie Xu, Shijie Li, Jianjun Li
Summary: Human activity recognition using wearable sensors has various applications but faces challenges such as incomplete feature extraction and low utilization rate of features. To address this, a novel deep multifeature extraction framework based on attention mechanism (DMEFAM) is proposed, achieving high recognition accuracies of 97.9%, 96.0%, and 99.2% on the WISDM, UCI-HAR, and DAAD datasets respectively, outperforming other advanced HAR frameworks.
IEEE SENSORS JOURNAL
(2023)
Article
Computer Science, Information Systems
Diana Nagpal, Shikha Gupta, Deepak Kumar, Zoltan Illes, Chaman Verma, Barnali Dey
Summary: Physical and mental health of elderly individuals can be improved with a Human Activity Recognition (HAR) system. We proposed a personalized feature fusion algorithm, goldenAGER, which extracts handcrafted HOG features and self-learned VGG-16 features to recognize abnormal activities. Our model achieved 95% accuracy on a primary dataset and 93.08% accuracy on the Microsoft Research (MSR) Action dataset, outperforming existing models.
Article
Engineering, Electrical & Electronic
Chaolei Han, Lei Zhang, Shige Xu, Xing Wang, Hao Wu, Aiguo Song
Summary: Deep convolutional networks have achieved great success in sensor-based human activity recognition. Most existing works focus on extracting multiscale activity features by increasing network depth or width, which is not suitable for resource-limited mobile devices. To address this issue, we propose a diverse-branch convolution (DBC) scheme, which strengthens the capacity of vanilla convolution by exploiting diverse branches of different scales and complexities. DBC can be transformed into a single convolution layer for activity recognition deployment while maintaining the model's performance and inference-time structure.
IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT
(2023)
Article
Engineering, Electrical & Electronic
Nabasmita Phukan, Shailesh Mohine, Achinta Mondal, M. Sabarimalai Manikandan, Ram Bilas Pachori
Summary: The variation in vital signs may be attributed to daily physical activities rather than organ defects. This study proposes a convolutional neural network-based human activity recognition method and evaluates its performance using acceleration signals from a standard benchmark database. The results demonstrate the importance of selecting optimal hyperparameters and the number of layers to achieve higher accuracy and shorter computational time.
IEEE SENSORS JOURNAL
(2022)
Article
Computer Science, Information Systems
Zhouping Chen, Jianyu Yang, Hualong Xie
Summary: In this paper, gesture recognition models based on sEMG were proposed. By constructing basic models and improving them using a multi-stream fusion strategy, high recognition accuracy was achieved. Experimental results showed that all models had high accuracy in gesture recognition and each had its own strengths.
Article
Computer Science, Artificial Intelligence
Beibei Zhang, Hongji Xu, Hailiang Xiong, Xiaojie Sun, Leixin Shi, Shidi Fan, Juan Li
Summary: This paper proposes a new activity recognition framework based on spatiotemporal multi-feature extraction with SCbSE blocks. By simulating the prison environment and collecting an aggressive activity dataset, a threshold-based aggressive activity detection method is developed to simplify the model and enhance recognition speed.
JOURNAL OF AMBIENT INTELLIGENCE AND HUMANIZED COMPUTING
(2021)
Article
Computer Science, Information Systems
Gangtao Han, Zheng Dong, Jian-Kang Zhang, Xiaomin Mu
Summary: In this letter, an ultra-reliable low-latency communication (URLLC) scheme for one uplink massive single-input multiple-output (SIMO) system with three users is considered. A new multiuser space-time modulation scheme is devised for all users to update status information to the base station (BS) concurrently with extremely low latency. Additionally, a noncoherent maximum likelihood (ML) receiver is investigated for the receiver side to detect all signals of different users simultaneously with high reliability and low latency when the antenna array size is scaled up. Extensive computer simulations are carried out to validate the proposed design's effectiveness in the case of a large antenna array size.
IEEE WIRELESS COMMUNICATIONS LETTERS
(2021)
Article
Computer Science, Information Systems
Wentong Han, Zheng Dong, He Chen, Xiangchuan Gao
Summary: This letter addresses the constellation design problem in energy-based noncoherent massive SIMO systems over correlated channels. An approximation method based on maximizing the minimum KL divergence of received signal vectors is proposed, and the difficult max-min KL divergence problem is optimally solved through Szego's theorem and a one-dimensional bisection search.
IEEE WIRELESS COMMUNICATIONS LETTERS
(2022)
Article
Computer Science, Information Systems
Congrui Fu, Hui Yuan, Hongji Xu, Hao Zhang, Liquan Shen
Summary: Light field technology can capture the four-dimensional information of light rays, including position, direction, and depth information. To improve the accuracy of depth estimation, a depth estimation algorithm based on convolutional neural network (CNN) is proposed. The algorithm utilizes a single image super-resolution algorithm to enhance the sub-aperture images and partitions the images into simple and complex texture regions based on texture analysis. Epipolar plane images (EPIs) are extracted and fed into specified network branches for both texture regions. A fusion module is used to generate the depth map. Experimental results show that the proposed method outperforms state-of-the-art methods in terms of objective and subjective quality, and is more robust to noise.
JOURNAL OF VISUAL COMMUNICATION AND IMAGE REPRESENTATION
(2023)
Article
Engineering, Electrical & Electronic
Yang Wang, Hongji Xu, Yunxia Liu, Mengmeng Wang, Yuhao Wang, Yang Yang, Shuang Zhou, Jiaqi Zeng, Jie Xu, Shijie Li, Jianjun Li
Summary: Human activity recognition using wearable sensors has various applications but faces challenges such as incomplete feature extraction and low utilization rate of features. To address this, a novel deep multifeature extraction framework based on attention mechanism (DMEFAM) is proposed, achieving high recognition accuracies of 97.9%, 96.0%, and 99.2% on the WISDM, UCI-HAR, and DAAD datasets respectively, outperforming other advanced HAR frameworks.
IEEE SENSORS JOURNAL
(2023)
Article
Engineering, Electrical & Electronic
Qian Zhang, Ju Liu, Zhichao Gao, Ziyu Li, Zhiying Peng, Zheng Dong, Hongji Xu
Summary: This paper proposes a robust transmission scheme for RIS-aided NOMA secure networks with transceiver HWI. A closed-form expression for the distortion noise power caused by transceiver HWI in NOMA networks is derived. The proposed scheme achieves more robust security and outperforms other networks without considering HWI and imperfect SIC.
IEEE TRANSACTIONS ON COMMUNICATIONS
(2023)
Article
Computer Science, Information Systems
Jie Xu, Hongji Xu, Shijie Li, Shuang Zhou, Mengmeng Wang, Yuhao Wang, Jiaqi Zeng, Jianjun Li, Xiaoman Li, Yiran Li, Xinya Li, Wentao Ai, Yang Wang
Summary: This paper introduces an algorithm for context inconsistency elimination based on a comprehensive measure of correctness and a two-dimensional mass function. The algorithm aims to solve the problem of context inconsistency in CASs and has been demonstrated to be effective through experimental analyses.
INFORMATION SCIENCES
(2023)
Article
Computer Science, Information Systems
Juan Li, Hongji Xu, Yuhao Wang
Summary: This article proposes a multiresolution fusion convolution network (MRFC-Net) to improve the accuracy of human activity recognition by correctly identifying confusing activities. It also introduces a multiresolution fusion convolution variational auto-encoder network (MRFC-VAE-Net) for open set HAR, which effectively classifies known and unknown class (UC) activities. Furthermore, a rich data set named daily-abnormal activity of special group (DAASG) is constructed for daily monitoring of special groups.
IEEE INTERNET OF THINGS JOURNAL
(2023)
Article
Computer Science, Information Systems
Yuhao Wang, Hongji Xu, Lina Zheng, Guozhen Zhao, Zhi Liu, Shuang Zhou, Mengmeng Wang, Jie Xu
Summary: In this study, a deep learning network for HAR based on MMS data is proposed, which fully utilizes the advantages of multidimensional convolutional kernels. Multiscale residual convolutional squeeze-and-excitation modules are also introduced to increase the diversity of feature information. The proposed network achieves high FW-scores on the PAMAP2 and OPPORTUNITY data sets using both tenfold and LOSO cross-validations.
IEEE INTERNET OF THINGS JOURNAL
(2023)
Proceedings Paper
Computer Science, Theory & Methods
Ran Yang, Ning Wei, Zheng Dong, Hongji Xu, Ju Liu
Summary: This paper proposes a robust reflection coefficient design for an IRS-aided system to enhance signal quality, addressing the imperfect adjustment of reflection coefficients. By linear approximation and alternating optimization methods, the non-convex optimization problem is converted into a sequence of convex subproblems for efficient solutions. Numerical results demonstrate that high resolution for phase shifts is not necessary for approaching ideal performance.
COMMUNICATIONS AND NETWORKING (CHINACOM 2021)
(2022)
